Runoff estimation method based on remote sensing data screening and error analysis

Abstract

The invention relates to the technical field of crossing of satellite gravity mechanics and hydrology, in particular to a runoff estimation method based on remote sensing data screening and error analysis, which comprises the following steps of: 1, constructing a novel remote sensing data screening method by fusing an effective pixel proportion and a cloud index method in a river reach range; the filtering module is used for filtering the data set to obtain a high-quality data subset and estimating the accuracy of model input parameters by runoff; 2, estimating the effective river reach width of a long-time sequence by utilizing a novel remote sensing data screening method and fusing a water body extraction method and a hydraulic geometrical relationship, and calculating the effective river reach width of the long-time sequence by combining a corrected grade curve method, an improved manning equation and a Bjerklie equation; the river runoff is estimated through the river surface hydraulic variable directly obtained from the remote sensing data set and the variable indirectly obtained from the empirical equation, the river runoff measurement information can be improved, and key data support is provided for more comprehensive understanding of water resource management and climate adaptation.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of satellite gravimetry and hydrology, specifically a runoff estimation method based on remote sensing data screening and error analysis that can improve the accuracy of runoff estimation. Background Technology

[0002] Rivers are complex natural water systems, serving not only as vital sources of water resources but also as crucial pathways for surface runoff to transport water bodies. Therefore, rivers provide essential life and ecological services to human activities and ecosystems. The components of river systems are an important part of the hydrological cycle, which forms the link between land and ocean reservoirs. River runoff is one of the important climatic variables in the water cycle, modulating the climate system and indicating its changes. Traditionally, river runoff has been continuously measured in situ at specific locations (i.e., using river level gauges). However, since 1980, the number of active gauges globally has declined sharply, with approximately 60% of publicly available gauge data being at least 10 years out of date. Furthermore, monitoring of small and medium-sized rivers is less effective, and river level gauges require instrument calibration and long-term maintenance, limiting their accessibility to physical proximity, largely confined to economically developed regions. The inability to obtain existing metrological data, particularly due to various economic and political reasons, exacerbates the global problem.

[0003] Significant efforts have been made to expand and extend traditional measurement methods to alleviate the scarcity of poorly measured or unmeasured watershed data. Remote sensing offers additional opportunities for acquiring runoff information, not only supplementing field data but also providing a truly global perspective on water resource availability. Summary of the Invention

[0004] This invention addresses the environmental challenges faced by the Yellow River Basin, particularly in the face of water scarcity, soil erosion, river pollution, and floods. By combining the modified order curve method, the improved Manning formula, and the Bjerklie equation, this invention proposes a runoff estimation method based on remote sensing data screening and error analysis. This method aims to improve the accuracy of runoff estimation and thus enhance water resource utilization and environmental conditions in the Yellow River region.

[0005] This invention achieves its purpose through the following measures: A runoff estimation method based on remote sensing data screening and error analysis, characterized by the following: Step 1: By integrating the effective pixel ratio and cloud index methods within the river section, a novel remote sensing data filtering method was constructed to filter the dataset to obtain a high-quality data subset, thereby improving the accuracy of the input parameters of the runoff estimation model. Step 2: Using a novel remote sensing data screening method and integrating water body extraction methods and hydraulic geometry, the effective river segment width for long-term series was estimated. By combining the modified order curve method, the improved Manning equation and Bjerklie equation, river runoff was estimated by combining the river surface hydraulic variables directly derived from the remote sensing dataset with the variables indirectly derived from empirical equations.

[0006] The novel remote sensing data screening method in step 1 of this invention specifically includes: First, remote sensing images significantly affected by clouds are removed using a cloud removal function, with the screening criterion being images with actual cloud cover below 20%; second, since the entire image after cloud removal often cannot represent the actual cloud cover within a small study area, a method is proposed to screen remote sensing data using the ratio between all effective pixels after cloud removal and all pixels within the study area, with the screening requirement being images with an effective pixel ratio greater than 0.90; finally, the river width of the long-term series is obtained using the high-quality remote sensing data set after screening as parameters for the runoff estimation model.

[0007] This invention also proposes a GRACE error analysis and evaluation method for screening novel remote sensing data, specifically: the water balance equation for a closed watershed is expressed as: (1) The runoff variation caused by different climatic factors is estimated by multiplying the changes of each climatic factor and its partial derivatives. Therefore, the multi-year average runoff depth variation is expressed in the following total differential form: (2), where, Indicates runoff depth (mm); Indicates potential evapotranspiration (mm); Indicates rainfall amount (mm); Indicates the change in water storage (mm); n This indicates underlying surface parameters (related to vegetation cover, topography, soil, vegetation, and other factors). Using formula (2) and the definition of elasticity coefficient, the elasticity coefficients of precipitation, potential evapotranspiration, underlying surface parameters, and water storage changes can be obtained. The calculation formulas are as follows: (3), in, , , and For the elasticity coefficients of changes in rainfall, potential evapotranspiration, underlying surface parameters, and water storage, where , representing the dryness index; In step 2 of this invention, based on data screening, an improved Normalized Difference Water Index (MNDWI) is proposed to quantify the degree of water inundation, as shown in the following formula: (4), of which, and These represent the reflectance of the green and mid-infrared bands, respectively, corresponding to bands B3 and B11 in Sentinel-2 data, corresponding to B3 and B6 in Landsat8 data, and corresponding to B2 and B5 in Landsat7 data. For Sentinel-1 satellite data, the Dual Polarization Water Index (SDWI) was selected to quantify the degree of water inundation. VV and VH Polarimetric image multiplication is used to enhance the contrast between water bodies and other surface features, thereby increasing the water body feature information and reducing the influence of other factors. The formula is as follows: (5) In order to better extract the local water surface of the river section, the water surface is binarized by Edge Otsu based on the water index method, so as to realize the water-land segmentation of the image and to calculate the water area; in order to obtain an accurate measurement of the river width, the water surface area of ​​the given river section is first calculated. The formula for calculating the water area is as follows: (6), in, S pi Area per unit pixel N This represents the total number of pixels on the river surface. S ri River surface area (m²) 2 ); The calculation of river width is accomplished by dividing the water surface area into selected appropriate river segments and choosing a relatively uniform river cross-section to avoid measurement errors in braided river segments. The formula is as follows: (7), Where W is the width of the river and L is the length of the river section (m), with L obtained using the Google Earth distance measurement tool.

[0008] Step 2 of this invention also includes the estimation of mechanical parameters, specifically: (1) Roughness coefficient estimation: The detailed process of using the estimated river roughness coefficient extraction method to distinguish river channel conditions through visual interpretation by Landsat satellite is as follows: (8), in, n 0 represents the basic roughness of a naturally straight, smooth, and uniform channel; n 1. The influence of irregular water surface; n 2. To take into account the influence of changes in the cross-sectional shape and size of the river channel; n 3. The influence of water-blocking materials;n 4. Impact of vegetation; n 5. The impact of river channel meandering; (2) Slope estimation: The river channel slope of three hydrological stations in the Yellow River Basin was estimated using STM DEM data with a resolution of 30 m. Based on the definition of slope, the appropriate length (RL) of the river segment was determined using a simple formula developed therefrom. (9), in, The minimum slope of the Yellow River section in Inner Mongolia Autonomous Region is represented by σ (3.1 cm / km), and the total height error value is 4.65 m. The slope is then calculated by defining the slope as the best-fit line of the river centerline elevation on RL. The river centerline elevation is defined and obtained using STRM DEM in ArcGIS. (3) River depth estimation: River depth estimation formula derived using the traditional runoff formula and Manning's formula: (10) in, For flow rate, For slope, n Roughness coefficient; The flow velocity can be estimated using the river's width and slope: (11).

[0009] This invention utilizes remotely sensed river width and inverses river runoff through the power-law relationship between the equation and river runoff (Formula 12): (12) in, Runoff (m³) 3 / s), The width of the river (m). a and b These are empirical parameters.

[0010] This invention also includes evaluating the model's performance using Nash and Sutcliffe coefficient (NSE), root mean square error (RMSE), relative root mean square error (RRMSE), mean absolute percentage difference (MAPD), and mean bias error (MBE) after inverting runoff using the model. (16) (17) (18) (19) (20) in, It is the measured runoff. It is an estimated flow rate. This represents the average value of the measured runoff. n This represents the total number of observations.

[0011] This invention estimates river runoff by leveraging the high correlation between satellite-derived river water information and field-measured flow rates, effectively filling gaps in runoff monitoring. Results show that the hydraulic geometry derived using a novel remote sensing data screening method can effectively capture river runoff—the results of all three methods show a strong correlation with field flow measurements (Nash efficiency coefficient (NSE) between 0.79 and 0.89, and relative root mean square error (RRMSE) between 9.81 and 31.10), demonstrating the potential of this novel remote sensing data screening method for rapidly and accurately estimating daily river runoff. The novel remote sensing data screening-error analysis assessment combined method proposed in this invention not only helps improve river runoff measurement information but also provides crucial data support for a more comprehensive understanding of water resource management and climate adaptation. Attached Figure Description

[0012] Appendix Figure 1 The geographical location and river channel view (a) of the Yellow River Basin mentioned in this invention and the Inner Mongolia Autonomous Region section of the Yellow River Basin (b) of the STRM-DEM digital elevation model.

[0013] Appendix Figure 2 This invention uses Landsat images to extract water body classification results, wherein... Figure 2 (a) is Sentinel 2 on June 2, 2020; (b) is Landsat 8 on July 10, 2020; (c) is Sentinel 1 on August 1, 2020.

[0014] Appendix Figure 3 This is a time series of the effective width of three research stations in the Inner Mongolia Autonomous Region section of the Yellow River Basin, derived from satellite imagery from 2013 to 2022.

[0015] Appendix Figure 4 The nominal curves for estimating the average flow depth and observed flow rate based on hydraulic geometry.

[0016] Appendix Figure 5The fitting curve between river width and measured runoff obtained from remote sensing satellite images, and the derived empirical parameters.

[0017] Appendix Figure 6 To compare the runoff results estimated during the validation period using the modified rank curve method with the measured runoff.

[0018] Appendix Figure 7 This is a comparison between the runoff results estimated during the verification period using the Bjerklie equation and the improved Manning formula in this embodiment of the invention and the measured runoff.

[0019] Appendix Figure 8 This describes the trend changes of various hydrological elements and the sudden changes in runoff in the embodiments of the present invention.

[0020] Appendix Figure 9 The monthly variation of the elasticity coefficients of various meteorological and hydrological elements in the embodiments of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] This example presents a method for estimating the Yellow River runoff based on remote sensing data screening and error analysis. The Yellow River has abundant water, but its precipitation distribution is uneven, with a north-south precipitation ratio greater than 5. Furthermore, some areas in Ningxia Hui Autonomous Region and Inner Mongolia Autonomous Region receive less than 150 mm of precipitation, facing severe water shortages. Considering the difficulty in monitoring runoff in plateau and mountainous areas, the Sanhuhekou, Toudaoguai, and Shizuishan sections of the Yellow River in Inner Mongolia Autonomous Region were selected as research sites. Figure 1 As shown, the Inner Mongolia section of the Yellow River is located at the upper reaches of the Yellow River basin, between 106°58′E and 111°10′E, and 40°15′N and 40°50′N. It is characterized by arid and semi-arid conditions, belonging to a temperate continental climate, with a multi-year average temperature of 9°C, large diurnal temperature range, and distinct seasonal variations. The Inner Mongolia section of the Yellow River features diverse natural landforms including grasslands, lakes, wetlands, and deserts. The land on both banks is fertile, making it an agro-pastoral transition zone. Annual rainfall varies extremely unevenly, decreasing from east to west, with approximately 75% of the rainfall concentrated between July and September.

[0023] The measured data comes from the daily measured runoff of Shizuishan, Toudaoguai, and Sanhuhekou from 2013 to 2022, and is used for parameter calibration of the rated curve and verification of the model inversion results. To eliminate the influence of ice conditions on model accuracy, this example only processes and analyzes the measured data from April to October. The corrected grade curve requires data calibration and verification. In this example, the data from 2013 to 2020 is the calibration period, and the data from 2021 to 2022 is the verification period.

[0024] Sentinel-1 / 2: Sentinel-1 consists of two satellites (Sentinel-1A and Sentinel-1B), launched by the European Space Agency on April 3, 2014 and April 25, 2016, respectively. They collect SAR images from a C-band Synthetic Aperture Radar (SAR) with different polarizations and resolutions. The radar provides single-polarization and dual-polarization options for vertical transmission with vertical receiving (VV) and vertical transmission with horizontal receiving (VH). To derive the backscattering coefficients, the Sentinel-1 dataset (Sentinel1A / 1B ground-range detection scenarios) provided by GEE has been preprocessed through the following steps: (1) application of orbit files; (2) thermal noise removal; (3) radiometric calibration; and (4) terrain correction.

[0025] Sentinel-2, a binary satellite constellation consisting of Sentinel-2A and Sentinel-2B, observes the Earth's surface with a 5-day repeat cycle. As part of the Copernicus Project, Sentinel-2 carried a Multispectral Imager (MSI), whose spatial resolution varies across different bands, ranging from 10 m (bands 2, 3, 4, and 8) to 20 m (bands 5, 6, 7, 8a, 11, and 12) and 60 m (bands 1, 9, and 10). This invention uses Level-1C products that have undergone radiometric, geometric, and orthorectified processing on a GEE platform for this study.

[0026] Landsat 7 / 8 Data: Landsat is a series of satellite programs jointly launched by NASA and the U.S. Geological Survey (USGS). Landsat 7 carries an Enhanced Thematic Mapper Plus (ETM+) remote sensing instrument with a resolution range of 15 m (band 8) to 30 m (bands 1, 2, 3, 4, 5, 6, 7, B9) and 60 m (band 6). Landsat 8 carries an OLI (Operational Land Imager) TIRS thermal infrared sensor with 11 bands. The spatial resolution varies between the bands, ranging from 15 m (band B8) to 30 m (bands B1, B2, B3, B4, B5, B6, B7, B9) and 100 m (bands B10, B11). On the GEE cloud platform, a series of preprocessing steps were performed on the Landsat data, including radiometric calibration, atmospheric correction, and geometric correction.

[0027] Supporting data: Monthly water storage variations were derived from the GRACE / GRACE-FO RL06 Mascon solution from the Center for Space Research at the University of Texas, Austin (CSR). The CSR RL06 mascon (CSR-M) is a 0.25° latitude and longitude grid, representing a 1° × 1° geodesic grid with equal area around the equator. Furthermore, no additional smoothing or filtering was applied to CSR-M. This invention uses the CSR-M solution from April 2002 to December 2020, but 33 months of data are missing; therefore, the missing data was estimated using cubic spline interpolation.

[0028] Monthly rainfall data comes from the Global Land Data Assimilation System (GLDAS), jointly developed by NASA and NOAA. This model utilizes surface modeling and data assimilation techniques to output various surface parameters (soil temperature, soil moisture, evapotranspiration, surface runoff, etc.) and hydrological variables. This invention uses the GLDAS Noah Land Surface Model V2.1 product (GLDAS-Noah), with a spatial resolution of 0.25°, and the data time span is consistent with GRACE.

[0029] Monthly evapotranspiration data are derived from potential evapotranspiration data from the GLEAM (Global Land Evaporation Amsterdam Model). GLEAM is an algorithm that estimates global-scale evapotranspiration by combining satellite observation data and utilizing the Priestley and Taylor equations. Since 2011, GLEAM has been continuously revised and updated, and the third version of the model was released in 2017.

[0030] This example demonstrates the construction of a novel joint method for remote sensing data screening and error analysis assessment. First, the novel remote sensing data screening method is introduced: Before data processing, filtering the dataset to obtain isomorphic subsets is crucial. Therefore, to obtain images with less cloud cover and higher image quality, a novel remote sensing data filtering method was constructed using the effective pixel ratio within the river segment and combined with the cloud index method. First, remote sensing images with significant cloud cover were removed based on a cloud removal function, with the filtering criterion being images with actual cloud cover below 20%. Second, since the entire image after cloud removal often does not represent the actual cloud cover within a small study area, a method was proposed to filter the remote sensing data using the ratio between all effective pixels after cloud removal and all pixels within the study area, with the filtering requirement being images with an effective pixel ratio greater than 0.90. Finally, the long-term river segment width was obtained from the filtered high-quality remote sensing dataset as a parameter for the runoff estimation model.

[0031] GRACE Error Analysis and Evaluation: River flow is influenced by various factors within the watershed, and this study analyzes and verifies these factors by considering evapotranspiration, rainfall, and changes in terrestrial water storage within the study area. By characterizing surface water resources using runoff and simulating watershed water resource evolution using water balance, the sources of error in runoff analysis are assessed. The water balance equation for a closed watershed is expressed as: (1), Changes in runoff caused by different climatic factors can be estimated by multiplying the changes in each climatic factor and their partial derivatives. Therefore, the multi-year average runoff depth variation can be expressed in the following total differential form: (2), where, Indicates runoff depth (mm); Indicates potential evapotranspiration (mm); Indicates rainfall amount (mm); Indicates the change in water storage (mm); n This indicates underlying surface parameters (related to vegetation cover, topography, soil, vegetation, and other factors).

[0032] Using formula (2) and the definition of elasticity coefficient, the elasticity coefficients of precipitation, potential evapotranspiration, underlying surface parameters, and water storage changes can be obtained. The calculation formulas are as follows: (3), in, , , and For the elasticity coefficients of changes in rainfall, potential evapotranspiration, underlying surface parameters, and water storage, where , which represents the dryness index.

[0033] Water extraction and river width calculation: Based on data screening, an improved Normalized Difference Water Index (MNDWI) is proposed to quantify the degree of water inundation. Compared with the Normalized Difference Water Index (NDWI), this method can effectively suppress interference from non-water body factors and extract water body information of rivers flowing through high-altitude areas. The formula is as follows: (4), of which, and These represent the reflectance of the green and mid-infrared bands, respectively, corresponding to bands B3 and B11 in Sentinel-2 data, corresponding to B3 and B6 in Landsat8 data, and corresponding to B2 and B5 in Landsat7 data.

[0034] For Sentinel-1 data, the Dual Polarization Water Index (SDWI) was selected to quantify the degree of water inundation. This method uses... VV and VH Polarimetric image multiplication enhances the contrast between water bodies and other surface features, thus strengthening water feature information and reducing the influence of other factors. The formula is as follows: (5), To better extract local water surfaces in river sections, this example combines the water index method with EdgeOtsu to binarize the water surface, thereby achieving land-water segmentation of the image for water area calculation. Figure 2 The image classifies Landsat images and uses a combination of image bands (visible and near-infrared bands) and Sentinel-1 dual-polarization bands to identify water pixels in the Inner Mongolia section of the Yellow River basin from other non-water pixels. Then, the water body index method and the maximum class variance threshold segmentation method are used to extract the degree of water inundation.

[0035] Various satellite sensors can directly measure river width at different temporal and spatial resolutions; however, this method is susceptible to errors due to vegetation, damp ground, and rocks obstructing the water's edge. Therefore, estimating river width using the water surface area digitized from the average river segment length in Landsat imagery is more accurate than direct measurement. Specifically, to obtain an accurate river width measurement, the water surface area of ​​a given river segment is first calculated using the following formula: (6), among which, S pi Area per unit pixel N This represents the total number of pixels on the river surface. S ri River surface area (m²) 2 ).

[0036] River width is calculated by dividing the water surface area into selected, appropriate river segments, and by choosing a relatively uniform river cross-section to avoid measurement errors in braided river segments. The formula is as follows: (7), where W is the width of the river and L is the length of the river section (m). L can be obtained by using the Google Earth distance measurement tool.

[0037] The mechanical parameters are estimated as follows: Roughness coefficient estimation: River channel resistance is significantly influenced by vegetation, materials involved, obstacles, meandering, and channel irregularity. The detailed process of differentiating channel conditions using visual interpretation from Landsat and extracting the estimated channel roughness coefficient is as follows: (8), among which, n0 represents the basic roughness of a naturally straight, smooth, and uniform channel; n 1. The influence of irregular water surface; n 2. To take into account the influence of changes in the cross-sectional shape and size of the river channel; n 3. The influence of water-blocking materials; n 4. Impact of vegetation; n 5. The influence of river channel meandering.

[0038] Slope estimation: To obtain reliable slope, this example uses 30 m resolution STM DEM data to estimate the channel slope at three hydrological stations in the Yellow River basin, and determines the appropriate length (RL) of the river segment using a simple relational formula developed based on the definition of slope. (9), among which, The minimum slope (3.1 cm / km) of the Yellow River section in Inner Mongolia Autonomous Region is represented by σ, which is the total height error (4.65 m). The slope is then calculated by defining the slope as the best-fit line of the river centerline elevation on RL. The defined river centerline elevation was obtained using STRM DEM in ArcGIS.

[0039] Sichangi et al. derived a river depth estimation formula using the traditional runoff formula and Manning's formula: (10), among which, For flow rate, For slope, n Roughness coefficient; The flow velocity can be estimated using the river's width and slope: (11).

[0040] Runoff refers to the amount of water passing through a cross section of a river per unit time, which can be calculated using water depth, flow velocity, and river width. Currently, there is no direct method for estimating river flow based on space; instead, it is calculated from observations of hydraulic flow parameters using mathematical formulas or calibration relationships. Estimating river runoff using the geometric relationship between the hydraulic geometry of the river cross section and runoff is a method adopted by many scholars. Therefore, this invention uses the river width extracted by remote sensing to invert the river runoff through the power-law relationship between the equation and the river runoff (Formula (12)).

[0041] (12), among which, Runoff (m³) 3 / s), The width of the river (m). a and b These are empirical parameters.

[0042] Discharge estimation based on statistically derived equations with accurate estimates of input parameters is also promising. Bjerklie et al. developed the hydraulic equation (equation (13)). Gleason used the same method to estimate river discharge from space and applied it to the Mekong and Ob rivers. Therefore, this example uses this formula to conduct an inversion study on the runoff of the Yellow River Basin (Bjerklie equation): (13); The runoff inversion method based on Manning's formula typically estimates the spatial discharge of a known ideal river cross-section using river system characteristics. This paper derives and develops the hydrological relationship (an improved Manning formula), proposing that when river width W >> water depth D, river runoff calculation can be expressed as: (14); This example details the surface hydraulic parameters obtained from remote sensing and their related hydraulic relationships, as well as the flow estimates obtained using the improved Manning formula and the Bjerklie equation, and compares their accuracy. It also calculates the normalized flow using formula (20).

[0043] (15), among which, To normalize the runoff, The runoff before normalization (including remote sensing flow estimation and on-site flow measurement). and These are the minimum and maximum measured runoff at a given study station during the study period, respectively. For example, during the study period, the runoff at the Sanhukou station... and Data were obtained on August 30, 2020, and July 30, 2017, respectively. The normalized runoff is in the range of 0 to 1.

[0044] After inverting runoff using the model, the performance of the model is evaluated using the Nash and Sutcliffe coefficient (NSE), root mean squared error (RMSE), relative root mean squared error (RRMSE), mean absolute percentage difference (MAPD), and mean bias error (MBE).

[0045] (16) (17) (18) (19) (20), among which, It is the measured runoff. It is an estimated flow rate. This represents the average value of the measured runoff. n This represents the total number of observations.

[0046] River width extraction results: The river water range of the Yellow River section in Inner Mongolia Autonomous Region was calculated using satellite remote sensing data measured within 10 km above and below three selected non-braided river sections on the GEE cloud platform. Figure 3 The effective width of the Yellow River section in Inner Mongolia Autonomous Region during the study period of 2013-2022 was estimated using formula (7). Figure 3 The figure shows the temporal variation of the watershed's hydrological wet and dry seasons and the density of collected data. During the study period (2013-2022), the estimated minimum, maximum, and average effective widths of the Sanhu River Estuary were 401.16 m, 758.71 m, and 568.93 m, respectively; the minimum, maximum, and average effective widths of the Toudaoguai River Estuary were 201.68 m, 331.01 m, and 266.66 m, respectively; and the minimum, maximum, and average effective widths of the Shizuishan River Estuary were 275.87 m, 434.63 m, and 356.40 m, respectively.

[0047] Average runoff depth estimation results: Since the Inner Mongolia section of the Yellow River Basin is mostly located in plateau areas, it is very difficult to monitor small (40–200 m wide) and medium (200–800 m wide) rivers by satellite altimetry. Therefore, since the Inner Mongolia section of the Yellow River Basin meets this benchmark, this invention uses formula (10) to calculate the average runoff depth. Therefore, this invention fits the estimated runoff depth to a power function relationship ( Figure 4 The correlations were found to be strong, with correlation coefficients of 0.903, 0.879, and 0.916, respectively. The average runoff depth estimates for the Sanhuhekou, Toudaoguai, and Shizuishan hydrological stations processed during the study period were 3.96 m, 2.7 m, and 3.47 m, respectively. The average depth estimation method used in this invention is feasible, especially for narrow mountain rivers such as those in the Inner Mongolia Autonomous Region section of the Yellow River basin.

[0048] Accuracy Evaluation of the Modified Rank Curve Method Inversion: The relationship between river width and discharge typically differs across different river cross-sections. Therefore, understanding and evaluating the hydraulic relationship between discharge and river width and water depth at different cross-sections is a crucial step in discharge estimation. This invention plots a nominal curve of the hydraulic relationship between the average river width obtained from hydrological stations using remote sensing and the actual runoff. Using measured data and river width data from 2013-2020 as calibration periods, the results show a strong correlation between river width and runoff, R0.2 All exceeded 0.85 (e.g. Figure 5 ).

[0049] Based on the power-law relationship between river width and river runoff, this invention uses parameters obtained periodically from the average river width and the pass rate of measured data at stations for inversion. The results show satisfactory agreement between the flow estimated using the relationship fitting method combined with satellite data and the observed flow in most river sections. At the Sanhukou station, the flow estimation showed an NSE of 0.89 during the validation period, 0.885 at Shizuishan, and 0.91 at Toudaoguai; the NSE coefficients of all three hydrological stations exceeded 0.85. The overall performance evaluation index NSE during the validation period indicates that the method of estimating river runoff using river width and measured runoff is reliable. Figure 6 It can be used for runoff estimation in the Yellow River basin.

[0050] The MBE coefficients at all three hydrological stations were less than 0, indicating that formula (9) has a lower overall prediction level. Shizuishan's NSE coefficient was lower than that of Sanhuhekou and Toudaoguai, while its RMSE performance was better. A reasonable explanation for this phenomenon is that, with the diversion of Yellow River tributaries, Shizuishan is located upstream of Sanhuhekou and Toudaoguai, resulting in significant runoff variations. For example, the average runoff in Shizuishan between 2021 and 2022 was 868 m³. 3 / s, the Sanhuhekou and Toudaoguai sections are 623 m apart. 3 / s, 645 m 3 / s.

[0051] according to Figure 6 It can be seen that the higher the correlation between river width and runoff, the better the estimation accuracy for the validation period, indicating that in the upper reaches of the Yellow River, river width can effectively reflect changes in river runoff. However, the limitation of using this method to estimate runoff is that the quality of optical images is easily affected by cloud cover and shadows. This is why the flow estimates for Sanhuhekou on June 5, 2022, Toudaoguai on June 13, 2022, and Shizuishan on September 6, 2022, have relatively large errors.

[0052] The estimated flow of the Yellow River in Inner Mongolia Autonomous Region was compared using the Bjerklie equation and the improved Manning formula, and their performance was verified by plotting. The model used the depth obtained from hydraulic parameters (10), the slope obtained from a 30 m resolution SRTM DEM, the width obtained from Landsat imagery, and the roughness coefficient obtained from the channel conditions derived from the references using the formula (8) to calculate the flow of the Lhasa River. Overall, the runoff estimation results of the Bjerklie equation and the improved Manning formula are consistent with the observed values ​​(e.g., Figure 7These two methods offer reasonable accuracy and can effectively capture changes in river runoff. However, compared to the modified ordinal curve method, the Bjerklie equation and the improved Manning formula, to some extent, overestimate runoff; the MBE (Mean Expectation) is greater than 0 and reaches a maximum of 197 m. 3 / s, the minimum is 42.01 m 3 Previous studies by Bjerklie et al., Sichangi et al., and Kebede et al. have mentioned that using models with multiple hydraulic flow parameters for flow estimation has provided feasible results for many rivers. The results of this invention are also applied to the Yellow River, a large river, and the analysis was performed on rivers in the Inner Mongolia Autonomous Region. The results show that the model can achieve better accuracy for larger rivers in the plateau region (NSE coefficient above 0.80, except for the worst case, the first channel flow estimated using Equation 10, with an NSE coefficient of 0.795).

[0053] The lower accuracy of the Bjerklie equation and the improved Manning formula for the Toudaoguai and Shizuishan hydrological stations is due to variations in river width and water level. Because of limitations in time-based data estimation, the estimated channel slope and roughness coefficients are assumed to be constant over time. While flow increases with river width and depth, the persistent overestimation of flow at these two stations is mainly due to the small variation in river width. Increased uncertainty in the estimation of river depth further contributes to the lower accuracy of flow estimation. However, for long-term and large-scale rivers, this error is tolerable. Figure 7 The estimated results are very close to the observed results, showing a strong positive correlation (the NSE coefficient is basically maintained above 0.79).

[0054] The Bjerklie equation and the improved Manning formula are multivariate statistical equations for flow estimation techniques, where performance varies from river to river. Equations (9) and (10) are applicable to flows below 20,000 m³. 3 The improved Manning formula performs well in rivers with flow rates of / s, exhibiting strong correlation and a high NSE value. Similarly, in this invention, the improved Manning formula performs well in estimating flow in large plateau rivers, showing the same consistency. However, due to the limited ability of current satellite sensors to accurately estimate certain flow parameters (such as identifying water bodies and riverbed slope), it is difficult to accurately estimate river flow at river cross-sections, which can introduce potential errors, thus interfering with and affecting the accuracy of flow estimation. Furthermore, variations in water and sediment, riverbed morphology, and the step of deriving slope ratings using a DEM in the Yellow River section of Inner Mongolia Autonomous Region may also introduce other errors.

[0055] For the Bjerklie equation and the improved Manning formula, considering all data within the study period (2013-2022), compared to the modified rank curve method which only considers data from the validation period (2021-2022), the overall performance of the inversion results relative to the measured data has decreased. The NSE value of Sanhuhekou changed from 0.892 in Model 1 to 0.882 and 0.887, and the RRMSE changed from 12.9% to 20.89% and 20.87%. The NSE value of Shizuishan changed from 0.885 in the modified rank curve method to 0.809 and 0.832, and the RRMSE changed from 9.81% to 23.2% and 21.52%. The NSE value of Toudaoguai changed from 0.908 in the modified rank curve method to 0.795 and 0.811, and the RRMSE changed from 13.4% to 31.1% and 29.71%. Typically, the overall error in estimating runoff using the Bjerklie equation and the modified Manning formula is caused by… S and n The errors introduced during the river estimation process are relatively minor. Therefore, for long-term use of remote sensing to calculate river width, the increase in outliers in the estimated river width using satellite sensors leads to a decrease in accuracy. However, overall, the runoff estimated by the three methods matches the measured runoff (e.g.,...). Figure 6 and Figure 7 This study demonstrates the feasibility of using three methods to retrieve runoff in the Inner Mongolia Autonomous Region section of the Yellow River Basin and the ability to monitor flow using satellites.

[0056] Runoff inversion error analysis: Trend changes of various hydrological elements, such as Figure 8 ,from Figure 8 (a) It can be seen that the fluctuation range of water storage is increasing, with the largest change occurring in January 2018 (-198.58 mm) and the smallest change occurring in August 2013 (6.07 mm). Figure 8 (d) Compared to this, it can be observed that the time of the largest change in water storage coincides with that of the largest change in runoff, with slopes of -0.013 and 0.002 respectively, indicating that changes in water storage have a significant impact on runoff variation. Combined with... Figure 8 As can be seen from (b) and (c), the potential evaporation and rainfall change with the same magnitude, and the maximum and minimum values ​​occur at the same time. They also maintain an upward trend with the change in runoff. However, the slope of rainfall (0.003) is greater than that of potential evaporation (0.0004), indicating that the runoff is affected by the synchronous effect of rainfall and evaporation to a certain extent, which causes errors in the runoff inversion of river parameters.

[0057] This example uses the Petitt test to assess the trend significance of runoff in the Inner Mongolia Autonomous Region of the Yellow River Basin from 2013 to 2020. The results show that the upward trend is significant at a confidence level of 0.05 (absolute value of statistic is 0.003), and the maximum statistical value is shown in 2017. Therefore, October 2017 was selected as the time point of abrupt change in runoff, with 2013-2017 as the baseline period and 2017-2020 as the period of change.

[0058] For the hydrological elements and elasticity coefficients of the Yellow River section in Inner Mongolia Autonomous Region, the average water storage, average precipitation, and average runoff during the change period increased compared to the baseline period, by 0.26%, 0.04%, and 119%, respectively. Based on the elasticity coefficients, during the change period... The value is the largest, followed by... and The smallest is This indicates that runoff is most sensitive to changes in precipitation, and only... The value is greater than zero and is 1.37, indicating that for every 10% increase in precipitation, runoff will increase by 137%, while , and The values ​​were -0.37, -0.45, and -0.11, respectively, indicating that for every 10% increase in potential evapotranspiration, underlying surface index, and terrestrial water storage, runoff would decrease by 37%, 45%, and 11%, respectively. Compared to the baseline period, The absolute value of [value] increased during the period of change, indicating that the effect of the underlying surface change on runoff change became more significant. , and The absolute value of the value decreased compared to the baseline period, indicating that the underlying surface and potential evapotranspiration had a smaller impact on runoff changes during the change period.

[0059] Figure 9 The changes in the elasticity coefficients of precipitation, potential evaporation, underlying surface index, and water storage from 1971 to 2009 show that... and The absolute value changes tend to be horizontal and the trends of both are consistent because... + =1, therefore the difference between their absolute values ​​is 1. The frequency of peak values ​​of the absolute value increased after October 2016, indicating that the effect of the underlying surface on runoff changes gradually increased. Corresponding to the previous analysis, the absolute value of the elasticity coefficient of the underlying surface during the influence period was slightly larger, but the node of occurrence was earlier than in the previous article. This may be due to the lag between changes in the underlying surface and changes in runoff. The large fluctuations indicate that significant changes in terrestrial water storage can greatly affect the hydrological cycle of the basin, causing abnormal runoff variations.

Claims

1. A runoff estimation method based on remote sensing data screening and error analysis, characterized in that, Includes the following: Step 1: By integrating the effective pixel ratio and cloud index methods within the river section, a novel remote sensing data filtering method was constructed to filter the dataset to obtain a high-quality data subset, thereby improving the accuracy of the input parameters of the runoff estimation model. Step 2: Using a novel remote sensing data screening method and integrating water body extraction methods and hydraulic geometry, the effective river segment width for long-term series was estimated. By combining the modified order curve method, the improved Manning equation and Bjerklie equation, river runoff was estimated by combining the river surface hydraulic variables directly derived from the remote sensing dataset with the variables indirectly derived from empirical equations.

2. The runoff estimation method based on remote sensing data screening and error analysis according to claim 1, characterized in that, The novel remote sensing data screening method in step 1 specifically includes the following steps: First, remote sensing images heavily affected by clouds are removed using a cloud removal function, with the screening criterion being images with actual cloud cover below 20%. Second, since the entire image after cloud removal often cannot represent the actual cloud cover within a small study area, a method is proposed to screen remote sensing data using the ratio between all effective pixels after cloud removal and all pixels within the study area, with the screening requirement being images with an effective pixel ratio greater than 0.

90. Finally, the river width of the long-term series is obtained using the high-quality remote sensing data set after screening as parameters for the runoff estimation model.

3. The runoff estimation method based on remote sensing data screening and error analysis according to claim 2, characterized in that, In step 2, based on data screening, an improved Normalized Difference Water Index (MNDWI) was proposed to quantify the degree of water inundation. The formula is as follows: （4）， in, and These represent the reflectance of the green and mid-infrared bands, respectively, corresponding to bands B3 and B11 in Sentinel-2 data, corresponding to B3 and B6 in Landsat8 data, and corresponding to B2 and B5 in Landsat7 data. For Sentinel-1 satellite data, the Dual Polarization Water Index (SDWI) was selected to quantify the degree of water inundation. VV and VH Polarimetric image multiplication is used to enhance the contrast between water bodies and other surface features, thereby increasing the water body feature information and reducing the influence of other factors. The formula is as follows: （5）， To better extract local water surface areas in river sections, the water surface is binarized using the Edge Otsu method in conjunction with the water index method, thereby achieving land-water segmentation of the image for water area calculation. To obtain accurate river width measurements, the water surface area of ​​a given river section is first calculated using the following formula: （6）， in, S pi Area per unit pixel N This represents the total number of pixels on the river surface. S ri River surface area (m²) 2 ); The calculation of river width is accomplished by dividing the water surface area into selected appropriate river segments and choosing a relatively uniform river cross-section to avoid measurement errors in braided river segments. The formula is as follows: （7）， Where W is the width of the river and L is the length of the river section (m), with L obtained using the Google Earth distance measurement tool.

4. The runoff estimation method based on remote sensing data screening and error analysis according to claim 3, characterized in that, Step 2 also includes the estimation of mechanical parameters, specifically: (1) Roughness coefficient estimation: The detailed process of using the estimated river roughness coefficient extraction method to distinguish river channel conditions through visual interpretation by Landsat satellite is as follows: （8）， in, n 0 represents the basic roughness of a naturally straight, smooth, and uniform channel; n 1. The influence of irregular water surface; n 2. To take into account the influence of changes in the cross-sectional shape and size of the river channel; n 3. The influence of water-blocking materials; n 4. Impact of vegetation; n 5. The impact of river channel meandering; (2) Slope estimation: The river channel slope of three hydrological stations in the Yellow River Basin was estimated using STM DEM data with a resolution of 30 m. Based on the definition of slope, the appropriate length (RL) of the river segment was determined using a simple formula developed therefrom. （9）， in, The minimum slope of the Yellow River section in Inner Mongolia Autonomous Region is 3.1 cm / km, and σ is the total height error value (4.65 m). The slope is then calculated by defining the slope as the best-fit line of the river centerline elevation on RL. The river centerline elevation is defined and obtained using STRM DEM in ArcGIS. (3) River depth estimation: River depth estimation formula derived using the traditional runoff formula and Manning's formula: （10）， Where V is the flow velocity and S is the slope. n Roughness coefficient; The flow velocity can be estimated using the river's width and slope: （11）。 5. The runoff estimation method based on remote sensing data screening and error analysis according to claim 1, characterized in that, Using the river width extracted by remote sensing, the river runoff is inverted through the power-law relationship between the equation and the river runoff, i.e., Formula 12: （12）， in, Runoff (m³) 3 / s), W is the width of the river (m). a and b These are empirical parameters.

6. The runoff estimation method based on remote sensing data screening and error analysis according to claim 5, characterized in that, This also includes evaluating the model's performance using Nash efficiency coefficient (NSE), root mean square error (RMSE), relative root mean square error (RRMSE), mean absolute percentage difference (MAPD), and mean bias error (MBE) after inverting runoff using the model. （16）， （17）， （18）， （19）， (20), among which, It is the measured runoff. It is an estimate of the flow rate. This represents the average value of the measured runoff. n This represents the total number of observations.

Images